ci(apple): characterize bounded xcrun stalls in iOS preflight and fold-helper gate

[thymikee]

Purpose

An Apple host stall can fail smoke before the test under review starts. Keep the two xcrun paths distinct so a cold simulator inventory failure is not reported as an XCTest assertion or hidden by a test retry.

Evidence and scope

• iOS PR run 36030060382, job 107736587593 failed in Preflight iOS runner through public CLI. The typed error was COMMAND_FAILED: xcrun timed out after 15000ms, with args: ["simctl", "list", "devices", "-j"], empty stdout/stderr, and timeoutMs: 15000. The fixture E2E step never ran. The 15 s budget is LIST_TIMEOUT_MS in packages/platform-apple/src/readiness/runtime.ts; this is distinct from the cold toolchain version probes closed by ios(ci): cold toolchain probes time out below the first-exec signature stall (10 s / 5 s budgets) #2422.
• The macOS fold-helper -Werror gate invokes xcrun with FOLD_HELPER_BUILD_TIMEOUT_MS = 30000 in packages/platform-apple/src/foldable/fold-helper-cache.test.ts. This is a separate compile path to audit. I have not established a recent failure of this gate from the currently visible macOS runs, so its budget should be changed only if a captured duration or timeout warrants it.

Required behavior

1. Attribute a preflight failure to the exact simctl operation, its elapsed time, remaining prepare deadline, and whether a retry could join/reuse a warmed result. A timeout must stay bounded and typed; no generic XCTest or E2E retry is the fix.
2. Reproduce or characterize the 15 s inventory stall on a cold/contended host and choose a bounded recovery using the remaining prepare budget. A warm second call may be tried only when the first typed timeout and deadline allow it. Verify the simulator is still the scoped target and that a persistent hang still fails promptly.
3. Measure the fold-helper warning gate's cold and warm xcrun duration separately. If it reaches its 30 s budget, file the concrete failing run here and fix that gate at its own compile seam while retaining -Werror; otherwise close that audit item without changing the timeout.

Completion

• A focused fake-host test plants a first inventory timeout followed by a valid scoped result, plus a persistent timeout, and proves both outcomes and total budget.
• A macOS gate measurement or failing job link settles the fold-helper audit.
• A live iOS PR run gets through preflight on the exact fixing head. Record runner queue and execution time separately.

Related: #2422 (closed toolchain identity probes), #2491 (rotating iOS smoke failures), #2879 (fold request envelope; a different budget).